|M.Sc Student||Brueller Raviv|
|Subject||Image Denoising via Optimal Projection and Shrinkage|
|Department||Department of Electrical Engineering||Supervisors||Professor Michael Elad|
|Professor Yacov Hel-Or|
|Full Thesis text|
The image denoising problem is well-known and extensively studied: given an arbitrary image degraded by noise, we would like to remove only the noise without sacrificing image content. In recent years shrinkage-based methods for image denoising have stimulated interest by suggesting a relatively simple yet effective framework for the task. In a typical shrinkage-based noise suppression procedure a transform is applied to the image, the derived values are modified, and then used to reconstruct the denoised image.
Our work promotes the development of a source separation algorithmic framework adapted to image denoising. The approach we take is in the spirit of the shrinkage technique. We address the challenge by attempting to find an optimal transform as a set of projections, and optimal modification operations as look-up tables. To this end, we utilize an algorithm to find the corresponding parameters in a sequential process of optimizations, based on a training mechanism. Then we develop a variation of this algorithm. This development incorporates into the framework the use of the maximal rejection classifier (MRC) and the slicing transform (SLT), two algebraic tools introduced in recent years. The method presented has several virtues that make it advantageous. Promising experimental results illustrate the potential of the proposed framework.